Ear Recognition Using Ensemble of Deep Features and Machine Learning Classifiers
Bilal Attallah, Youcef Brik, Cheyma Nadir · 2022
Ear recognition is the most recent field of study for biometrics. Over the past two decades, feature-based ear identification methods and traditional machine learning classifiers have been the most frequently employed methodologies. Convolutional Neural Networks are currently being used in ear recognition studies to extract features. CNNs are capable of learning more precise characteristics robust to huge image variations and achieving state-of-the-art identification performance. In this research, ear recognition models created using CNN features are presented and evaluated. Our study can be divided into two stages. To overcome the insufficient training samples, we first apply data augmentation using several geometrical techniques. Secondly, the feature extraction and classification tasks are performed by the three CNN algorithms, after which we use MobileNetV2 to extract the features and SVMs and K-nearest neighbors to classify them to confirm the individual’s identification. The performance of the suggested model is tested and assessed using the AMI dataset. With MobilenetV2 and Support Vector Machine, the accuracy of our suggested technique was 99.33%. According to experimental findings, the proposed system performs well when compared to current approaches.